Lagging Ahead: Why Your Leading Indicators Are Trailing the Market

Most boardrooms still draw a clean line between leading indicators and lagging ones. Lagging metrics are the financial outputs that tell you what already happened: revenue, EBIT, gross profit, utilisation. Leading metrics are supposed to tell you what is about to happen: customer satisfaction scores, pipeline conversion, repeat purchase rates, all the rest. The first set is for the post-mortem. The second set is for the strategy.

There is a problem with that distinction, and it is becoming more pronounced as more businesses run themselves on the back of dashboards. The leading indicators are also lagging. They are dressed up as forward-looking but they are made of the same substance as the financials, which is historic data, just sampled earlier in the chain. They tell you what is about to happen only insofar as the past three months reliably predict the next three. In most categories, in most markets, that reliability is much weaker than the dashboards imply.

If your leading metrics need three months of data before they show a shift, you do not have a leading metric. You have a lagging metric with a smaller delay. The lag is shorter than the financial lag, but the substance is the same. It is information about what already happened, not insight into what is about to happen.

This matters because the distinction we have lost is not the distinction between leading and lagging. It is the distinction between data and information.

The Quiet Inversion

For most of business history, leading indicators were not metrics at all. They were judgements. A senior salesperson reading a market told you what was about to happen because they had spent twenty years reading markets and they could feel the change before any number would show it. A senior product person told you which features mattered because they had watched customers use the last ten products and they knew what the next ones needed to do. A senior operator told you which contracts were going to fall over because they had seen the warning signs in a hundred previous deals.

This kind of leading indicator was informational. It compressed a vast amount of experience into a judgement that arrived before the data did, by definition, because the person was running the prediction in their head from a model that had been built over decades. It was leading in the proper sense of the word. It got there first.

What we now call leading indicators is something different. They are correlations. We have learned that certain measurable inputs statistically precede certain measurable outputs, and we have started to call those inputs leading. They are leading only in a thin, time-series sense. They sit earlier on the clock than the outcomes they predict. That is all the word means now. The deep predictive capacity that used to live in the heads of experienced people has been quietly replaced by historic data sampled at a faster interval. The word stayed the same. The thing it pointed at changed completely.

You can see this shift most clearly in how decisions get justified in modern boardrooms. The justifications that carry weight are the ones backed by numbers. The justifications that get treated as anecdote are the ones backed by judgement. A senior person saying “this will not work” is asked for evidence. The evidence is required to be quantitative. If they cannot produce a chart that shows their concern, the concern does not survive the meeting. So we run the experiment, generate the data, and act on what the data says, which is by then a description of a fact that has already crystallised.

An accountancy firm I worked with not long ago illustrates this almost perfectly. They had moved, with the best of intentions, to a more commercial footing. Revenue, EBIT, utilisation and gross profit became the focal points of every senior meeting. The dashboards were clean. The numbers were rigorous. The cadence was disciplined. And in the rear-view mirror they could see exactly where the problems had been four to six months earlier, which by the time anyone noticed had already worked their way through into the financials. Every quarterly review was a sophisticated analysis of issues that had already happened and could no longer be addressed except by reaction. The leading bit, the bit that would have told them which service lines were drifting, which clients were quietly disengaging, which kinds of accountancy practice they were actually becoming versus the one they wanted to be, was nowhere in the metrics. It was in the heads of the partners who had stopped being asked.

The dashboards are clean. The cadence is disciplined. The decisions are four months late. Lagging ahead is still lagging.

Four Reasons We Inverted

There are, I think, four reasons this inversion has happened, and they reinforce each other in ways that make the trajectory hard to reverse.

The first is defensive. Quantitative justifications are easier to defend in a post-mortem than qualitative ones. If a decision based on data goes wrong, the data can be blamed. If a decision based on judgement goes wrong, the person who made the call carries the failure personally. As organisations have become more risk-averse and more litigious, decision-makers have shifted toward justifications that distribute accountability. Data does that. Judgement does not. This is not the noblest reason, but it is real and it explains a lot of what looks like rigour but is actually cover.

The second is a loss of trust in expertise. Too many confident experts have been confidently wrong, in public, in this century, for the cultural authority of expert judgement to survive intact. Markets, governments and consumers have grown sceptical of the senior voice in the room. Data feels neutral by comparison. It does not have a personality, a bias, or a career to protect. The shift to data-driven decision-making is partly an attempt to find a substrate for decisions that does not depend on people we have decided not to trust.

The third is technological. The rise of machine learning and predictive analytics has changed what we mean by “leading indicator” in a way that hides the change. The new meaning is correlational. The old meaning was predictive in the strong sense, based on causal understanding of a system by someone who knew it well. We have collapsed these into a single phrase and lost the distinction without noticing.

The fourth, and the one I think does the most damage in practice, is an obsession with scale.

The Scale Trap

Until the late twentieth century, scale meant physical reach. Ships, factories, distribution networks, decades of compounding capital. The companies that achieved scale did so over long horizons, and their playbooks were built around the constraints of physical commerce: relationships, reputation, repeat customers, local presence, the slow accumulation of trust.

The dotcom era changed what scale meant. Scale became a software problem. The playbooks for achieving it changed accordingly, and the new playbook was data-driven, because at sufficient scale the data is right often enough to be operationally useful even when it is wrong on any individual case. Amazon can lose a particular customer to Etsy or a boutique retailer and not care, because Amazon is playing a probability game across hundreds of millions of customers. The data is good enough at that volume. The losses do not aggregate to anything that matters at the level Amazon operates.

The problem is that almost no business operates at Amazon’s scale, and the playbook does not transfer cleanly downward. A boutique retailer using data-driven decision-making in the same way Amazon does loses twenty-five percent of its addressable customers to better-targeted alternatives and goes under, because at boutique scale there is no probabilistic floor. You do not have enough customer interactions for the data to average out. You need to be right about the customer in front of you, and the customer in front of you does not show up in the data until after they have left.

This is the bit that I think is genuinely underappreciated. Data-driven decision-making at Amazon’s scale is rational. At small and mid-market scale it is often a category error, because the conditions that make large-N statistical thinking work do not apply. You are running a probabilistic playbook on a deterministic problem. The customer you lost did not get averaged out across a million similar customers. They were a meaningful percentage of your business and they are not coming back, and the data that would have told you they were leaving arrived three weeks after they left.

For almost every business below a certain scale, the right playbook is the older one. Know your customers individually. Read the market through people who understand it. Treat data as confirmation rather than direction. The Amazon playbook is not generalisable, and trying to apply it to a business that does not have Amazon’s scale produces the worst of both worlds: the slow, expensive infrastructure of a data-driven approach with none of the probabilistic averaging that makes it work.

The Jane Street Exception

There is one industry that has spent more money than any other trying to drive the lag out of data-driven decision-making, and it is worth pausing on because it proves the argument rather than refuting it.

High-frequency trading firms like Jane Street operate on the only timescale at which the distinction between leading and lagging genuinely collapses. They have driven the latency between market event and trading response down to microseconds, with billions of dollars in compute, custom silicon, fibre optic cable routed in straight lines between exchanges, and some of the most expensive engineering and mathematical talent on earth. At that frequency, “historic data” means the market state from a few microseconds ago, which is close enough to the present that the lag stops being decision-relevant for the kinds of bets they are making.

This is the limit case where data-driven actually does what the rhetoric claims. And it is the limit case for a reason. Even with all that investment, the data still lags. Jane Street is not predicting the future. They are reacting faster than anyone else to a present that has just become past. The system works because the lag has been driven below the threshold at which it matters for their specific game, not because the lag has been eliminated.

Everyone else is playing the same game on a slower clock. A boardroom looking at last quarter’s revenue is reacting to a present that became past three months ago. A dashboard looking at last week’s churn is reacting to a present that became past seven days ago. The data is always historic. The question is whether your latency is small enough relative to the speed of your market that the historicity stops mattering. For Jane Street, sometimes yes. For almost everyone else, no.

This is the trap of looking at firms like Jane Street and concluding that data-driven decision-making works. It works for them because they have spent a generation of effort and capital to push the lag below the threshold of their specific decisions. The lesson is not that data-driven approaches work. The lesson is that data-driven approaches only work when the lag is small relative to the timescale of the decision, and the lag has been brought down to the point where the system has paid the price of reducing it. Most businesses have done neither. They have adopted the rhetoric of data-driven decision-making without the latency to make it work and without the scale to make probabilistic averaging save them. They are playing Jane Street’s game on a four-month delay.

What We Lost When We Lost the Expert

The conventional way to describe the loss of expert judgement is to say there are fewer senior voices in the room, or that they have less authority when they speak. This is true, but it understates the loss by treating expertise as a credential rather than a capability.

The thing an expert actually carries is a working internal model of a category, built over twenty or thirty years of close observation. The model is mostly tacit. They cannot articulate most of what is in it. When a senior person says “this will not land” or “this is the moment to push,” they are running a simulation in a model they could not fully describe even if you asked them to, and the output of the simulation is a judgement that arrives faster than the explanation for it.

What this means is that the expert is doing exactly what the data-driven approach claims to do. They are taking historic patterns and projecting them forward. They are just doing it in a substrate that has been trained on far richer data than any spreadsheet captures. Body language at past meetings. The smell of a launch that was overconfident. The way a junior team behaved before the last failure. Customer conversations from fifteen years ago that taught them what people say when they are about to leave versus when they are about to buy. None of this is in any dataset. All of it is in the model. The expert’s judgement is not anti-data. It is compressed-data, run through a model that the data-driven approach does not have access to.

When we devalue this kind of judgement in favour of dashboards, we are not removing a flawed input in favour of a more reliable one. We are removing the only system in the building that can integrate qualitative signals into a predictive output, and replacing it with a system that can only see what has already been quantified. The data system cannot do what the expert does, because the qualitative signals were never quantified in the first place. The junior analyst cannot do it because the model takes decades to build. And no process or framework can replace the expert, because the expertise was never legible enough to be processed or framed.

There is a second loss that compounds the first, and it is the one that makes the trajectory hard to reverse. When experts are devalued, they stop being asked. When they stop being asked, the apprentices who would have learned from them stop forming. The model that took thirty years to build does not get transferred, because the transfer mechanism was always implicit: years of watching an expert work, absorbing their judgements, internalising the shape of their model. Devalue the expert in this generation and you do not just lose this generation’s experts. You lose the next two generations as well, because there is nobody to apprentice under. By the time the limitations of data-driven decision-making become obvious in your business, the people who could have helped you do anything else will have retired, and you will not have built their replacements.

This is connected to a related point I have made elsewhere about junior professionals and cognitive overload. The long, slow, unmeasurable work of building expert models has been crowded out by systems that prefer legible, measurable, immediate output. Two generations of business decisions have been made on the assumption that this trade is worth it. The bill for that trade is not yet visible on most income statements. It will become visible when the experts who are still around finish retiring.

What Finance Is Actually For

If data and financial modelling are not the right basis for the decision, what are they for?

They are for removing blockers. A decision is justified upstream of the numbers, in the qualitative reading of the market by people who understand it. The marketing case, the product case, the operations case, the customer case. Finance’s job is to check that the action you have decided to take is affordable, sequenceable, and survivable in the downside. It is not to tell you what to do. By the time finance is the deciding voice in the room, you have already conceded that you do not know what to do and you are hoping the numbers will tell you.

They will not. They cannot. Numbers are not that kind of object. They are descriptions of the past, projected into the future on the assumption that the future will resemble the past closely enough for the projection to be useful. In a stable system this assumption is roughly true. In a competitive market, in a shifting category, in any domain where the interesting decisions are being made, the assumption is false often enough to make data-led decisions worse than expert-led ones over time.

The right division of labour is this. Marketing, product, operations and leadership identify what the business should do, based on their qualitative understanding of the market. Finance checks that the proposed action is feasible and sustainable. The numbers are gatekeepers, not authors. When the numbers are the authors, you are letting the past write the future, and the past does not know what is coming.

The Practical Position

For a senior leader trying to act on any of this, four questions are worth separating.

First, what is your scale, and is your scale actually large enough for a data-driven playbook to do what it claims to do? Most businesses are not at that scale, and they are paying for the infrastructure of data-driven decision-making without getting the probabilistic averaging that makes it pay off. Be honest about which playbook you are running and whether you have the volume to make it work.

Second, is your latency from data to decision small enough that the data is still meaningful when you act on it? If your dashboards refresh weekly but your market moves daily, you are making last week’s decisions. The fix is not better dashboards. The fix is either reducing the latency, which is usually expensive and often impossible, or accepting that data is not the right input for your speed of decision and rebuilding the qualitative capacity to act on faster signals.

Third, who in your business is carrying the expert model for your category, and are you listening to them? If the answer is “nobody,” that is the most important strategic problem in your business, and it is invisible on every dashboard you currently look at. If the answer is “this person, but we treat their judgement as anecdote until the data confirms it,” you are making decisions four months too late and calling it rigour.

Fourth, what is the role of your finance function in your decision-making, and are they being asked to author decisions or to gatekeeper them? If your CFO is the loudest voice in your strategy meetings, your strategy is going to be a description of the past dressed up as a plan for the future. That is not a CFO problem. It is a leadership problem. The remedy is not to weaken finance. It is to strengthen the qualitative voices around it so the numbers are the check rather than the source.

Most of the businesses I see making this mistake do not know they are making it. They believe they are being rigorous because they are being quantitative. The dashboards are clean. The cadence is disciplined. The decisions are defensible. And the decisions are also, in most cases, four months late and reactive, because the data the decisions are based on is by definition historic and the leading indicators have been silently replaced with faster-sampled lagging ones.

Lagging ahead is still lagging. Knowing the difference is what senior leadership is for.

---

If your business is making decisions on the back of dashboards and you are starting to wonder whether the data is showing you what you need to see, talk to us. Esbee’s management consultancy work helps senior leaders rebuild the role of qualitative judgement in decisions that have become quietly data-led, and design operating models that put finance back in its proper seat: as the check on a decision, not the author of it.